In the first part of this series, I outlined my thesis that 100% rule based trading can have multiple advantages for investors, like the accountability of past performance and the lack of room for second thoughts; which poses a huge benefit for individuals who tend to doubt their own decisions and have problems sticking to their trading plan.
Additionally, system trading can contribute to the overall stability of an equity portfolio by breaking down correlations to other asset classes.
To validate those points, I try to build a very simple and easy to trade system, by bundling a few statistical edges and adding the necessary risk and money management routines that shall prevent too deep losses.
The first exploitable edge introduced was the bullish bias on Fed days and this article will add the remarkable bullishness on the first trading days of the month.
Part 2: First day of the month
I heard of that phenomenon several times over the last decade, usually explained with mutual fund and 401k inflows that have to be deployed in the market at the beginning of the month, although I never really bought that explanation and therefore ignored it all together.
But once I tracked this claim with statistical evidence, it became pretty obvious that for whatever reason this particular bullish bias indeed exists.
Here are the definitions of the examined setup:
- We buy the S&P 500 on the close of the last trading day of any given month, hold the position for exactly one day and sell on the close of the first trading day of the following month.
Our test period stretches from March 1st 1988 until March 1st 2013, or exactly 25 years.
No stop orders, slippage or transaction fees will be considered
|Trades total:||300||Days total:||6297|
|Winning Trades:||183||(61%)||Up Days:||3372||(53.5%)|
|Losing Trades:||117||(39%)||Down Days:||2925||(46.5%)|
|Avg. Winner:||+0.909%||Avg. Profit:||+0.758%|
|Avg. Loser:||-0.837%||Avg. Loss:||-0.8%|
|Max. Winner:||+4.0%||Max. Profit:||+11.58%|
|Max. Loser:||-8.93%||Max. Loss:||-9.03%|
A hit rate of over 60% with the average profit exceeding the average loss suggests a credible edge and is a good base for a profitable system.
To see if this edge might strengthen in bullish market phases compared to bearish ones, we can filter the results by adding the criteria that the last close of the month should be above the 200 day moving average in order for us to go long.
If Close above 200 SMA
|Trades total:||217||Days total:||4511|
|Winning Trades:||137||(63.1%)||Up Days:||2435||(54%)|
|Losing Trades:||80||(36.9%)||Down Days:||2076||(46%)|
|Avg. Winner:||+0.785%||Avg. Profit:||+0.606%|
|Avg. Loser:||-0.538%||Avg. Loss:||-0.632%|
|Max. Winner:||+3.125%||Max. Profit:||+5.12%|
|Max. Loser:||-2.642%||Max. Loss:||-6.87%|
The slight increase of the hit rate is nothing to get excited about, but the big decrease in the average losing trade and the maximum losing trade shows that we filtered out some of the worst stretches.
So even though we reduced the total trade number by over 25%, an increased expectancy lifts the total performance close to the original, unfiltered setup and makes for a smoother equity curve.
Being exposed to any market risk without protection spells disaster, especially when aiming for such specific edges, which is why a hard stop is a no-brainer in this case:
So let's add the same simple and fix 1.5% stop loss that we used in the first part of this series.
|System unfiltered||System filtered|
|Trades total:||300||Trades total:||217|
|Winning Trades:||183||(61%)||Winning Trades:||137||(63.1%)|
|Losing Trades:||117||(39%)||Losing Trades:||80||(36.9%)|
|Avg. Winner:||+0.909%||Avg. Winner:||+0.785%|
|Avg. Loser:||-0.608%||Avg. Loser:||-0.479%|
|Max. Winner:||+4.0%||Max. Winner:||+3.125%|
|Max. Loser:||-1.5%||Max. Loser:||-1.5%|
It shows that this obvious step adds a lot of stability to both versions by lowering the avg. loss and therefore raising the expectancy.
The fact that the hit rates stay the same, shows that we did not run into big reversal days (getting stopped out at -1.5%, before the market reverses and ends higher in the end). This speaks for the presented edge and one could experiment with a tighter protection, but we absolutely don´t want to over-optimize our setup, which is why I will stick with a 1.5% stop loss.
The one result that really stands out is the way better overall performance of the unfiltered original setup, due to the higher trade frequency with about the same expectancy as the filtered version.
One could be tempted to right away declare the version with the higher performance to be the better one, but let us have one more cautious look at another popular risk metric: the experienced drawdowns.
As expected, the 200 sma filter does a good job in sorting the worst phases out; with only 4 drawdowns over 2% and the maximum just under 4%, compared to 7 occasions of over 2% and two drawdowns of up to 6% in the unfiltered version.
However, trying to determine a better version is close to splitting hair and also depends on personal preferences and the final money management routine of the system.
So for the sake of non-over-optimization and a constant trade frequency (the filtered version was inactive for 19 months during 2007-2009 and only traded once in 31 months during the 2000-2003 bear market!), I will use the original version without factoring the 200 day moving average in.
With this second exploited seasonality we bring the activity of the overall system to 20 days a year, so about one tenth of the possible exposure.
In the third part of this series I will introduce another simple edge and later apply a set of routines that will define the necessary money management and tell us when to abandon a setup completely, in order to round up a complete rule based system.